Automated Assistant Multilingual Intent Recognition via Canonical Translation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional automated assistants are resource-intensive to configure for communication in new languages, requiring generation of new trigger grammars, intent recognition, and natural language output, and often lack expertise to handle language idiosyncrasies, limiting their ability to understand and respond to queries in non-native languages.
Innovation Solution
The use of a machine translation model that maps semantically-related natural language inputs to canonical translations, optimized for intent determination, allowing automated assistants to interpret and respond to queries in multiple languages by training a sequence-to-sequence model to map multiple phrases to a single canonical translation, reducing the need for extensive language-specific components and expertise.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If conventional techniques are used to configure automated assistants for new languages, then language-specific components (trigger grammars, intent recognition, natural language processing) can be generated, but the resource consumption and complexity increase significantly
Solution Approach 1:
The patent applies universality by using a single machine translation model to handle multiple languages instead of creating separate language-specific processing components. The model translates queries from various languages into a canonical language, allowing the automated assistant to serve multiple languages through a unified processing pipeline, thereby reducing system complexity while expanding language coverage
Solution Approach 2:
The patent introduces a machine translation model as an intermediary component between the user's native language and the automated assistant's processing language. This mediator translates diverse language inputs into a canonical form that the assistant can handle, eliminating the need for direct language-specific components for each supported language and reducing overall system complexity
2Reliability
If language-specific natural language processing components are created for each language, then intent recognition accuracy improves, but the time and resources required for system configuration increase
Solution Approach 1:
The patent applies preliminary action by pre-training the machine translation model on parallel corpora and language-specific nuances before deployment. This pre-processing ensures that when the system encounters queries in non-native languages, the translation model is already optimized for accurate intent preservation, eliminating the need for time-consuming configuration of language-specific NLP components for each new language
Solution Approach 2:
The patent changes the approach from creating language-specific processing parameters to using a single translation model with adjustable translation quality parameters. By optimizing translation parameters and using techniques like back-translation and domain adaptation, the system maintains high intent recognition accuracy across multiple languages without requiring separate NLP component configurations
3Adaptability or versatility
If multiple language-specific NLP components are implemented, then the automated assistant can handle more languages, but the computational resources required increase
Solution Approach 1:
The patent merges multiple language-specific NLP processing paths into a single machine translation-based processing pipeline. Instead of running separate intent recognition and processing systems for each language, the solution combines them into one unified flow where translation handles the language-specific adaptation, significantly reducing computational resource consumption while maintaining support for multiple languages
Solution Approach 2:
The patent extracts the language-specific processing requirements from the core automated assistant system and delegates them to the machine translation model. By separating the translation function from the intent recognition and execution components, the system reduces its own computational burden while still achieving multilingual capability through the external translation layer
Data Source
Figure 1
Figure 2
Figure 3
AI summary
Techniques described herein may serve to increase the language coverage of an automated assistant system, i.e. they may serve to increase the number of queries in one or more non-native languages for which the automated assistant is able to deliver reasonable responses. For example, techniques are described herein for training and utilizing a machine translation model to map a plurality of semantically-related natural language inputs in one language to one or more canonical translations in another language. In various implementations, the canonical translations may be selected and/or optimized for determining an intent of the speaker by the automated assistant, so that one or more responsive actions can be performed based on the speaker's intent. Put another way, the canonical translations may be specifically formatted for indicating the intent of the speaker to the automated assistant.